Abstract

Kavita Bala: Scalable illumination and textures for high-complexity scenes

In this talk I will present our latest research on scalable rendering for high-complexity scenes. Scene complexity continues to grow in computer graphics, driven by the availability of acquisition devices such as scanners and cameras. Applications such as massively-multiplayer games (MMPs), movies, surgical training, architectural renderings, all require robust algorithms for rendering high-complexity scenes. However, these applications currently must restrict themselves to low-quality rendering models and simple scene representations. Additionally, these applications currently require extensive manual intervention to achieve their goals. Scalable, robust illumination algorithms are required for the complex scenes rendered in these applications. While geometric level-of-detail algorithms abound, level-of-detail algorithms for illumination in complex scenes have not been explored.

In this talk I will describe our recent research on scalable illumination for high-quality rendering. Multi-dimensional lightcuts is a scalable rendering algorithm for high-quality illumination. This approach efficiently supports disparate rendering effects such as motion blur, global illumination, and depth of field, in one unified framework. The cost of state-of-the-art techniques is linear in the number samples required to compute each effect. In contrast, multidimensional lightcuts is sublinear (nearly constant), thus achieving 15x speedup and better quality over the fastest alternative (Metropolis).

I will also briefly describe other research from my group on interactive cinematic relighting, and texture representation and synthesis. I will conclude with a discussion of future plans in several areas of interest: scalable rendering, parallelizing graphics for multicore technologies, and rendering to heterogeneous displays.

Joint work with: Adam Arbree, Milos Hasan, Ganesh Ramanarayanan, Bruce Walter